Resolution except for minimal singularities. I (Q715225)

The article under review deals with the following problem: what kind of partial resolution of the singularities of an algebraic variety \(X\) can we get if we do not modify points at which \(X\) has normal crossings? Of course, the notion of ``normal crossing'' must be clarified. Let us say that \(X\) has \textit{strict normal crossings} (snc) at a point \(x\) if, locally near \(x\), there is an embedding of \(X\) into a regular variety \(U\) so that \(X\) is locally defined by a monomial expression involving a regular system of parameters of \({\mathcal O}_{U,x}\). We say that \textit{X has normal crossing at \(x\)} (nc) if there is an etale neighborhood \((X',x')\) of \((X,x)\) so that \(X'\) has simple normal crossings at \(x'\). We let \(X^{snc}\) and \(X^{nc}\) denote the set of points of \(X\) having snc and nc respectively. Working throughout over a field of characteristic zero, the authors prove that, given an algebraic variety \(X\), there is a projective morphism \(f:X' \to X\), such has the only singularities of \(X'\) are snc points, and \(f\) induces an isomorphism from \(f^{-1}(X^{snc})\) onto \(X^{snc}\). The morphism \(f\) is a composition of blowing-ups with admissible centers, which can be chosen in a ``canonical'' or ``algorithmic'' way. A similar statement involving normal crossings rather that simple normal crossings is not true. If we avoid modifying nc points of \(X\) may be we cannot get a partial resolution whose only singularities are nc points. The ``Whitney umbrella'' defined by \(z^2-xy^2=0\) in \({\mathbf C}^3\) is a counter-example. So, a natural question (posed by Kollar) is: what singularities are unavoidable when we use centers that do not include nc points? The problem is easily reduced to the special case where \(X\) is a hypersurface in a regular scheme \(Z\) of dimension \(n\). The authors solve this problem for \(n \leq 4\). For instance, identifying singularities which become isomorphic on a suitable etale neighborhood, for \(n=3\) the complete list \(\mathcal S\) of particularly simple representatives of the resulting equivalence classes, called normal forms, consists of those defined, in local coordinates \(x, y, z\) by \(xy\), \(xyz\) or \(z^2 +xy^2\) (a pinch point.) They also find, for \(n \leq 4\), the smallest class \(\mathcal S'\) of singularities, containing regular and nc points, such that by suitably modifying a variety \(X\) with centers that avoid points in \(\mathcal S'\) we get a variety \(X'\) whose only singularities are in \(S'\). For \(n \leq 3\), \(\mathcal S = \mathcal S '\), for \(n >3\) the set \(\mathcal S '\) is strictly larger. The tools used in the proofs come from the authors' fundamental article on resolution [Invent. Math. 128, No. 2, 207--302 (1997; Zbl 0896.14006)]. They involve Hilbert functions, auxiliary marked ideals and, specially, the numerical invariant inv, used to control the resolution process. Another feature is a \textit{Cleaning Lemma}, a simplification technique resembling the one used to resolve monoidal ideals in the basic papers on desingularization. From properties of inv it follows that the resolution sequences obtained are functorial with respect to local isomorphisms and etale maps. The authors do not include proofs for the case \(n=4\), they will appear in a sequel, written in cooperation with P. Lairez. They make some comments on the general case. The paper contains several interesting examples and an appendix where the authors review their fundamental resolution technique, specially the invariant inv.